Pembuatan Sketsa dari Foto Objek Nyata Menggunakan Generative Adversarial Network dan Deep Reinforcement Learning

Rahmayanti, Shintya Rezky (2021) Pembuatan Sketsa dari Foto Objek Nyata Menggunakan Generative Adversarial Network dan Deep Reinforcement Learning. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

Text 05111740000017-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2023. Download (2MB) | Request a copy

Abstract

Teknologi robot dan machine learning telah banyak diterapkan di berbagai bidang, tidak terkecuali bidang seni. Robot Paul mampu menggambar sketsa wajah manusia menggunakan metode filter konvolusi konvensional. Metode Generative Adversarial Network (GAN) telah berhasil dalam pembuatan gambar sintetis. Penelitian dalam pembuatan sketsa telah dilakukan baik menggunakan Recurrent Neural Network (RNN) maupun Deep Reinforcement Learning dengan mekanisme pembuatan goresan secara bertahap.
Pada penelitian ini akan dibuat sistem pembuat sketsa dari foto objek nyata menggunakan GAN dan Deep Reinforcement Learning. Adapun kerangka kerja yang digunakan mengacu pada Doodle-SDQ (Doodle with Stroke Demonstration and Deep Q-Network) yang menggabungkan proses pelatihan menggunakan metode supervised learning dilanjutkan dengan reinforcement learning. Foto objek nyata dengan kanal warna RGB dikonversi menjadi gambar kontur menggunakan GAN yang kemudian menjadi gambar referensi dalam pembuatan sketsa tiruan oleh agen reinforcement learning.
Pengujian dilakukan dengan penggantian pooling layer pada tahap supervised learning dan skenario rare exploration pada tahap reinforcement learning. Hasil penelitian ini mampu mencapai rata-rata total reward 2558.95 dengan rata-rata error piksel 0.0439 dengan step maksimum 200 dalam waktu rata-rata pembuatan sketsa 3.29 detik.
=====================================================================================================

Technology in Robotics and machine learning have been applied in numerous fields including arts. Paul The Robot is able to draw sketch from human face using conventional convolution filter method. Generative Adversarial Network (GAN) has been successful in generating synthetic images. Researches in sketch generation have been conducted either by using Recurrent Neural Network (RNN) or by using Deep Reinforcement Learning, with step-by-step stroke drawing.
In this research, we propose a system to generate sketch from real object images using GAN dan Deep Reinforcement Learning. The framework used is based on Doodle-SDQ (Doodle with Stroke Demonstration and Deep Q-Network) that combines training stages using supervised learning and is continued by the training stage using reinforcement learning. Real object images with color channel RGB are converted into contour drawing by GAN and then they will be the reference images by the reinforcement learning agent to generate the sketch.
The testing is done by modifying pooling layers during the supervised learning stage and rare exploration scenarios during the reinforcement learning stage. The result of this research is a model that can reach an average total reward of 2558.98 with an average pixel error of 0.0489 using 200 as the maximum step in an average time of 3.29 seconds in generating each sketch.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	sketch generation, generative adversarial network, deep reinforcement learning, pembuatan sketsa, generative adversarial network, deep reinforcement learning.
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Shintya Rezky Rahmayanti
Date Deposited:	13 Aug 2021 07:44
Last Modified:	13 Aug 2021 07:44
URI:	http://repository.its.ac.id/id/eprint/86226

Actions (login required)

View Item